The framebuffer size for fractional scale surfaces should be rounded to
the closest integer, not truncated as is currently done.
This implements the rounding formula that the discussions over at
wayland-protocols item 132 and MR 309 appear to be converging on.
Fixes#2713Closes#2810
Libdecor is a thin wrapper around xdg_toplevel, and expects the same
kind of information passed to it as if it was a xdg_toplevel. This
includes the min/max size when non-resizable, which was set when using
xdg_toplevel directly, but not when using libdecor. This fixes issues
with min/max size missing when mapping a non-resizable window.
Related to #2842
In order to safely call libdecor_frame_set_visibility during window
creation, a libdecor frame state was synthesized and committed.
This commit instead postpones the call to libdecor_frame_set_visibility
until after the frame state received by the frame configure event has
been committed. This appears to be a more polite approach.
Related to #2842
If there was already an event available in the Wayland queue, that event
would be processed but other kinds of events (libdecor, key repeat,
joystick connection, etc.) would not be.
This commit restores the processing of all available events on every
call to glfw*Events. It is partially based on #2842 by jadahl.
Because some libdecor plugins also process non-Wayland events, without
providing an fd for them, libdecor_dispatch is now called
unconditionally.
Fixes#2793Closes#2795
Related to #2842
This fixes a re-definition warning for the _GNU_SOURCE macro.
Edited by elmindreda: code formatting, author credit
Related to go-gl/glfw#359
Fixes#2833
If the frame was processed by wl_display_dispatch_queue_pending, this
would be treated as no frame having been received.
Oops.
This bug was introduced by fdd14e65b1.
When the swap interval was nonzero, EGL on Wayland would wait
indefinitely for a frame callback on a suspended window, causing
applications that call eglSwapBuffers to halt.
This re-implements the wait for a surface frame so that we can add
a reasonable timeout, while setting the EGL swap interval to zero. That
allows applications to make progress despite the window being suspended.
Some insights for how to structure this workaround were gleaned from
dri2_wl_swap_buffers_with_damage in Mesa.
This initial implementation still has a race between glfwHideWindow and
glfwSwapBuffers when rendering on a separate thread. This could lead to
a window becoming visible again after being hidden on the main thread.
Related to #1350Fixes#2582Fixes#2640Fixes#2719Fixes#2723Fixes#2800Fixes#2827
This adds support for high-resolution scroll input scaled to match
legacy discrete scroll input. It is available on compositors that
support wl_seat version 8 or later.
High-resolution "discrete" values now replace non-scaled scroll
values when available.
This adds support for wl_seat version 5, which adds pointer frames.
This allows two-dimensional scroll input to be emitted as single events
instead of one event per axis per motion.
It's also the the foundation for future enhancements like properly
scaled discrete scroll events and server-side key repeat.
Fixes#2494
The client-side key repeat timer fd was only created for wl_seat version
4 or later, but was then used unconditionally during event processing.
Rather than have this mechanism do nothing in a more correct way on
wl_seat version 3 or earlier (which is very old by now), this commit
creates the key repeat timer fd and then sets (hopefully gentle)
hardcoded repeat delay and rate.
Fallback decoration mouse button actions were performed for both press
and release events. The requests made using a mouse button release
serials were presumably (and correctly) discarded by the compositor.
This commit removes the generation of these nuisance requests.
If a modal surface like the window menu was active, clicking on the GLFW
window content area to close it would correctly emit the cursor enter
event but would not propagate the cursor position from the event.
When a click through to the fallback decorations caused the end of
a modal like the window menu, the cursor shape would not be updated
until the next time the cursor moved.
This commit adds an update of the cursor for the pointer enter event for
fallback decoration surfaces, in addition to the updates at pointer
motion events.
The fallback decorations would place the menu at the wrong position, by
not translating the last decoration surface position into toplevel
surface coordinates.
This also limits the menu to the caption area of the top decoration
surface, similar to how other toolkits work.
If fallback decorations were in use, pointer motion over a decoration
surface would cause glfwGetCursorPos to provide incorrect cursor
positions.
The cursor position is now only updated when the pointer is over the
content area of the window, similar to libdecor and XDG decorations.
Windows with keyboard focus may have an active key repeat timer.
This should be reset when the window is closed, or key repeat events
could be sent to a NULL window were it not for the quickfix in PR #2732.
Fixes#2741
Probably the source of #2727
libxkbcommon already provides functions to convert keysyms to codepoints
and UTF-8. The library has offered these functions since 0.5.0[1], so
using them won't cause any compatibility problems.
[1] https://xkbcommon.org/doc/0.5.0/group__keysyms.htmlCloses#2444
This adds the GLFW_SCALE_FRAMEBUFFER window hint, enabling control of
framebuffer scaling across Wayland and macOS. On macOS, this window
hint is a new name for GLFW_COCOA_RETINA_FRAMEBUFFER, and both hint
names will modify the same hint.
This is now a more symmetric counterpart to GLFW_SCALE_TO_MONITOR and,
weirdly, they each apply neatly to half of the supported platforms.
This commit is mostly documentation updates to better integrate and
contrast these two scaling mechanisms.
Camilla Löwy